The New Startup Formula: AI-First, Agency-Driven, Donkeycorn-Minded
The age of slow and expensive intelligence — humans — is over. For the rest of our lives we'll be living in the world of fast, cheap and powerful intelligence. Here's what it means for founders.
This is the talk I gave this week at an AI conference organised by Foundrs. Many people told me they found it useful, so I’m sharing it here.
We’re living through a profound shift in how startups are built. Every single person in our generation learned to build companies in the era of slow and expensive intelligence. By that, I mean human intelligence: smart, but costly, limited in working hours, and slow to scale. That kind of intelligence shaped everything we know about building companies—raising capital, managing teams, finding product-market fit, scaling operations. All of it was shaped by the fundamental constraints of human bandwidth.
Now, for the rest of our lives, we’ll be building companies in an era of fast, cheap, and powerful intelligence. Intelligence that is already smarter than us in some ways—and certainly faster and cheaper. The entire entrepreneurial playbook is going to be rewritten. The AI-first startup will be to today’s tech companies what Airbnb is to Hilton, or what Uber is to Addison Lee. Not just more digital, but fundamentally different.
So my intention is to explore what it means to build AI-first startups, what common mistakes to avoid, how to spot good opportunities, why donkeycorns might be more interesting than unicorns, and why, in the age of AI, human agency will matter more than intelligence.
Tech-First vs Tech-Supported
Let’s start not with AI, but with a classic case study: Netflix vs Blockbuster.
In the late ’90s, both companies were mailing DVDs to customers. You’d watch a film, send it back, and get a new one. Exciting times. But Netflix made a crucial bet: they saw broadband internet improving rapidly and imagined a world where people would stream movies on handheld devices instead of renting DVDs. They started building a streaming business while the infrastructure was still catching up. They let technological progress become a tailwind.
Blockbuster, on the other hand, used technology to optimise what they were already doing. They didn’t reinvent their business. So when broadband made DVD rentals obsolete, they were left behind. By the time they tried to launch a streaming service, Netflix had already won.
Netflix was tech-first. Blockbuster was tech-supported. That distinction is everything.
AI-First: The New Tech-First
Now, apply that same lens to AI-first startups.
The defining quality of an AI-first startup is that it’s enabled by and benefits from progress in AI. AI isn’t just a tool—it’s the foundation. The product gets better as AI gets better. It couldn’t exist without AI.
There are three additional characteristics that matter:
1. They deploy many AI agents
2. They learn continuously as they operate
3. They leverage human wisdom selectively
Let’s unpack these briefly.
Agents Everywhere
We already live in the era of AI agents. We can ask a tool like Deep Research to gather insights, summarise competitive updates, or automate multi-step workflows. Soon, tools like Manus and Operator will make it easy to deploy swarms of agents in production-grade systems. AI-first startups will have more agents than humans. Every employee will know how to configure, deploy, and orchestrate AI agents—just as we learned to hire and manage people.
Learning at the Speed of AI
While all companies learn over time, AI-first startups learn orders of magnitude faster. Humans are slow, reluctant learners. We’re attached to habits and comfort. AI, by contrast, is a voracious learner. Every customer conversation, every support ticket, every email becomes a datapoint in an ongoing feedback loop. Learning becomes ambient, automatic, and compounding.
Human Wisdom, Not Labour
Lastly, the best AI-first startups will still rely on human wisdom—the ability to discern what matters and why. Not human labour, not scale-by-hiring, but applied judgment. I can make a longer philosophical argument for this, but the short version is this: wisdom will be the limiting reagent in an economy where intelligence is abundant.
You Have to Use AI to Understand It
This might all sound obvious, but I regularly meet smart people who don’t get it. Not because they have considered objections—those are welcome—but because they glance at AI and dismiss it as “just another chatbot,” or the next cloud or crypto.
The mistake? They try to fit AI into their existing worldview, rather than update their worldview through use. I know because I made the same mistake with crypto. I evaluated it through familiar categories—money, ledgers, smart contracts, store of value—and found it lacking. But people who actually used crypto day-to-day saw it differently. They built new mental models. They got rich. I didn’t.
Even software developers—who arguably understand AI better than most—are often in denial. Their income and identity are wrapped up in a mental model of software development that AI now challenges. As Upton Sinclair said: “It is difficult to get a man to understand something when his salary depends upon his not understanding it.”
Many developers believe, “Writing code is just one part of the job; real software delivery is about systems thinking, communication, stakeholder management.” That’s true—but it’s also a comfort blanket. They don’t feel urgency to learn how to build software the AI-first way. There’s not enough market pressure yet. Their employers aren’t demanding it. But make no mistake: the tide is coming.
You are not qualified to have an opinion on AI unless you use it daily, no matter how smart or experienced you are.
For developers, for founders, for everyone—the right question is not “How do I use AI to improve what I’m already doing?” That’s incrementalism.
The right question is: What would my product or company look like if I built it from scratch with AI at its core?
That’s what Netflix did. That’s what Blockbuster failed to do. It’s what will separate tomorrow’s survivors from tomorrow’s punchlines.
Learn by Doing
You don’t update your worldview just by reading blog posts or attending talks. You do it by using AI every day. That’s how I learned, when I was helping a customer last week build an AI-first product. I spent hours on Upwork and Sales Navigator looking for a specific kind of freelancer—until I remembered to use Deep Research, which found the right person, who started the next day. That kind of tool-use reshapes how you think.
Another time, I used Deep Research to explore AI disruption opportunities. I formulated a simple thesis: small UK companies with proven product-market fit and human-intensive operations are vulnerable to AI disruption. Why invent something new when you can rebuild something that already works—but with AI, faster and cheaper?
One example it found was a reference-checking business turning over £10M. Their main cost? Humans calling schools and employers to verify credentials. Voice AI could automate much of that. Imagine undercutting their price by 70%, cutting costs by 90%, and still pulling in £2–3M of free cash flow. That’s the opportunity AI-first thinking reveals.
Enter the Donkeycorn
This brings me to one of my favourite hot topics: the donkeycorn.
Unicorns are out. Donkeycorns are in.
Donkeycorns grind like donkeys but party like unicorns. They’re small, highly profitable businesses with low headcount, no VC funding, and minimal complexity. Think: £2M ARR with 90% margins, run by a two-person team.
Historically, donkeycorns were rare because building them was hard. You needed time, capital, engineers. AI changes all of that. AI-first tooling lets you go from idea to working prototype in days. You can test for PMF quickly, get to revenue fast, and avoid the need for funding entirely.
Take Fyxer. They eventually took VC funding, but they got to $5M in ARR in 10 months—without it. They could have stayed a donkeycorn and printed money. That path is now available to more founders than ever.
Of course, there’s still a place for venture capital. But the equation has changed. It’s not the only playbook anymore.
Agency Is the Edge
So where does that leave us, as humans?
In a world of abundant, cheap intelligence, our edge will be agency. The capacity to choose goals, take initiative, adapt. In the industrial era, people traded agency for stability. They got jobs. Someone else made the decisions. But in the AI era, that model is fragile.
People often compare AI to the Industrial Revolution, but I think a better parallel is the collapse of the Soviet Union. I remember it as a child. When the planned economy collapsed, skills that once guaranteed a salary became useless overnight. It was brutal. But those who adapted—who had entrepreneurial instincts and a sense of agency—had a better shot at thriving.
That’s where we are now.
We’ll still need intelligence, but agency—our ability to act, to decide, to build—will matter even more.
TL;DR
If there’s one thing to take away, it’s this:
• Build AI-first: your product should benefit from and be enabled by AI progress
• Use AI every day to reshape your worldview
• Forget VCs—unless you truly need them
• Think donkeycorns: low-risk, high-margin, highly satisfying businesses
• And remember: agency is your edge. Be fast, be ready, and don’t wait for permission.